I'm a Masters' student in the Department of Electrical and Computer Engineering at Carnegie Mellon University with a concentration in AI/ML Systems. I'm working in the LIONS research group led by Dr. Carlee Joe-Wong. My research interest broadly lies in efficient machine learning and federated learning. I completed my B.Tech in Electrical and Electronics Engineering from NIT Andhra Pradesh, India, where I worked on several hands-on projects with a focus on deep learning and reinforcement learning to explore the field of AI. Prior to joining CMU, I worked as a Systems Engineer in Research at Tata Consultancy Services - Research where I spearheaded the design and optimization of various novel ultra wideband antennas and modeled ML solutions to enhance the usability of Reconfigurable Intelligent Surfaces with novel unit cell designs.
Under review
Yi Hu, I-Cheng Lin, Jinhang Zuo, Bob Iannucci, Carlee Joe-Wong, Ritvika Sonawane
ICDCS 2025 / [pdf]
Tapas Chakravarty, Poornima Surojia, Ritvika Sonawane, Sai Sarath Chandra Chaitanya Sayinedi, Meda Lakshmi Narayana, Soumya Chakravarty, Rowdra Ghatak
Patent Pending: US 2024/0372255A1 / [pdf]
Amartya Banerjee, Soumya Chakravarty, Ritvika Sonawane, Poornima Surojia, Tapas Chakravarty, Rowdra Ghatak
Patent Pending: US 2024/0364007A1 / [pdf]
Amartya Banerjee, Soumya Chakravarty, Ritvika Sonawane, Poornima Surojia, Tapas Chakravarty, Rowdra Ghatak
APSCON 2024 / [pdf]
Soumya Chakravarty, Poornima Surojia, Ritvika Sonawane, Tapas Chakravarty, Achanna Anil Kumar, Rowdra Ghatak
MAPCON 2023 / [pdf]
This project involved developing a personalized movie recommendation system using collaborative filtering (SVD) to enhance search relevance and user engagement. The system integrated a backend with SQL for data management and an API for seamless interaction. To ensure reliability, it incorporated MLOps best practices like model monitoring, A/B testing, and automated updates. Deployed with Docker and a load balancer, the system maintained scalability and 100% uptime.
This project focuses on developing a scalable search ranking system for streaming content, optimizing query relevance using TF-IDF, BM25, and BERT embeddings. The system improves content discovery by integrating collaborative filtering and content-based embeddings, leading to a 12% increase in NDCG score. A Flask API enables real-time ranked search queries, demonstrating efficient search optimization for streaming platforms.
This project investigates locally connected layers in speech models to enhance robustness and efficiency. Using untied convolutional kernels in the Conformer architecture on the LibriSpeech dataset, initial experiments showed minimal WER change (~2.3243) over 10 epochs. However, replacing convolutional subsampling layers with locally connected layers improved WER to 2.01 in just 5 epochs, suggesting better feature representation and faster learning. Future work focuses on optimizing hyperparameters to balance accuracy and computational cost.
Spring 2025, Carnegie Mellon University
Fall 2024, Carnegie Mellon University
Spring 2024, Carnegie Mellon University